Pre-trained random forest models obtained using preciseTAD
An experimentdata package to supplement the preciseTAD package containing pre-trained models and the variable importances of each genomic annotation used to build the model parsed into list objects and available in ExperimentHub. In total, preciseTADhub provides access to n=84 random forest classification models optimized to predict TAD/chromatin loop boundary regions and stored as .RDS files. The value, n, comes from the fact that we considered l=2 cell lines {GM12878, K562}, g=2 ground truth boundaries {Arrowhead, Peakachu}, and c=21 autosomal chromosomes {CHR1, CHR2, ..., CHR22} (omitting CHR9). Furthermore, each object is itself a two-item list containing: (1) the model object, and (2) the variable importances for CTCF, RAD21, SMC3, and ZNF143 used to predict boundary regions. Each model is trained via a "holdout" strategy, in which data from chromosomes {CHR1, CHR2, ..., CHRi-1, CHRi+1, ..., CHR22} were used to build the model and the ith chromosome was reserved for testing. See https://doi.org/10.1101/2020.09.03.282186 for more detail on the model building strategy.
